Skip to content

Design Bootstrap Cases (Common Real Tasks) #23

@jharris1679

Description

@jharris1679

Overview

Design bootstrap cases for common, everyday coding tasks applied to the user's actual repository. These test practical agent utility on real work.

Philosophy

Bootstrap cases must:

  • Work on any well-structured codebase
  • Respect project style and conventions
  • Produce minimal, targeted changes
  • Not break existing tests

Proposed Bootstrap Cases

ID Task Validation
boot-001 Fix all linting errors in a specific file Lint passes, tests pass
boot-002 Rename a symbol across the codebase Grep confirms, tests pass
boot-003 Extract duplicated code into shared function Pattern gone, tests pass
boot-004 Add type annotations to a module Type-checker passes, tests pass
boot-005 Update deprecated API usage No deprecated patterns, tests pass

Key Design Decisions

  1. Dynamic detection: Cases detect issues at runtime, not static examples
  2. Template prompts: ${VARIABLE} syntax for repo-specific context
  3. Minimal change validation: Track files changed, penalize over-modification
  4. Non-regression required: Tests must still pass

Tasks

  • Finalize 5 bootstrap case types
  • Design detection mechanism for each
  • Create prompt templates
  • Design "targeted change" evaluator
  • Implement case YAML files

Acceptance Criteria

  • 5 bootstrap cases with detection logic
  • Runtime template filling works
  • "Minimal change" scoring implemented
  • All cases validate non-regression

Metadata

Metadata

Assignees

No one assigned

    Labels

    documentationImprovements or additions to documentationenhancementNew feature or requestgood first issueGood for newcomers

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions